Large-scale Cellular Imaging of Neuronal Activity: a Study of Neural Individuality and a Method for Imaging Mouse Cortex

(1)

Washington University in St. Louis

Washington University Open Scholarship

All Theses and Dissertations (ETDs)

Summer 9-1-2014

Large-scale Cellular Imaging of Neuronal Activity:

a Study of Neural Individuality and a Method for

Imaging Mouse Cortex

Pei Sabrina Xu

Washington University in St. Louis

Follow this and additional works at:https://openscholarship.wustl.edu/etd

This Dissertation is brought to you for free and open access by Washington University Open Scholarship. It has been accepted for inclusion in All Theses and Dissertations (ETDs) by an authorized administrator of Washington University Open Scholarship. For more information, please contact

[email protected].

Recommended Citation

Xu, Pei Sabrina, "Large-scale Cellular Imaging of Neuronal Activity: a Study of Neural Individuality and a Method for Imaging Mouse Cortex" (2014).All Theses and Dissertations (ETDs). 1368.

(2)

WASHINGTON UNIVERSITY IN ST. LOUIS Division of Biology and Biomedical Sciences

Neurosciences

Dissertation Examination Committee: Timothy E. Holy, Chair

Yehuda Ben-Shahar Joseph P. Culver

Erik D. Herzog Karen L. O’Malley

Paul H. Taghert

Large-scale Cellular Imaging of Neuronal Activity: a Study of Neural Individuality and

a Method for Imaging Mouse Cortex

by Pei Sabrina Xu

A dissertation presented to the Graduate School of Arts and Sciences

of Washington University in partial fulfillment of the degree

of Doctor of Philosophy

August 2014 St. Louis, Missouri

(3)

c

(4)

LIST OF FIGURES

Figure Page

1.1 Pheromone-sensing neurons. . . 29 1.2 Pheromone-sensing neurons send information to the sexually dimorphic nuclei

in the brain. . . 31 1.3 Schematic of axial scanning and horizontal scanning. . . 37 3.1 Volumetric scanning of whole-mount VNO by OCPI. . . 51 3.2 Large-scale recording of vomeronasal sensory responses revealed non-stochastic

variability between individual mice. . . 54 3.3 Classifying specific VSN functional types using multiple stimulus responses:

an example of two steroid enantiomers. . . 55 3.4 Large-scale recording of sensory responses to sulfated steroids in male and

female mouse vomeronasal organs.. . . 58 3.5 Calcium signals at the dendritic knobs and the cell bodies were consistent

when whole-mount VNO was stimulated by sulfated steroids. . . 59 3.6 Physiological types of VSNs in single male and female imaging volume. . . . 60 3.7 Physiological neuronal types in VNOs from male and female mice. . . 62 3.8 A clustering-free analysis shows that epitestosterone sulfate (A6940)-selective

VSNs were specific to male mice. . . 63 3.9 VSN responsiveness to concentration series of sulfated steroids and female

urine tested by GCaMP3 VNOs. . . 67 3.10 Variability of isolation-housed male and female mice. . . 68 3.11 Exposure to female scents triggers the disappearance of a male-specific VSN

type. . . 70 3.12 Social exposure does not dramatically change the steroid response profiles of

VSNs in female mice. . . 71 3.13 Effects of sensory experience on males were seen only via decomposition by

VSN type, not on a per-ligand basis. . . 72 3.14 A clustering-free analysis shows that male mice exposed to female lost

(9)

Figure Page

3.16 The time window for experience-dependent plasticity of type-8 VSNs. . . 79

3.17 Investigation time of mice to sulfated steroids at different time intervals. . . 82

3.18 Olfactory investigation of epitestosterone was sexually dimorphic and deter-mined by experience. . . 83

3.19 Spatial distribution of cell types in the vomeronasal epithelium. . . 87

4.1 Schematic of hsOCPI setup. . . 96

4.2 Light sheet thickness. . . 97

4.3 Scanning the entire zebrafish nervous system. . . 100

(10)

ACKNOWLEDGMENTS

A PhD that easily takes us six years in the gold time between twenty to thirty is one of our most important decisions in the whole life. Now when I look back my past six years, I feel fortunate about my decision of moving here and joining in this program and this laboratory to pursue my PhD. It brings me to the fascinating sciences as well as happy life on each single day even when somethings go awry. I know that is because of a group of wonderful people that I meet here, and I sincerely thank them for all their help.

First, I appreciate the patient guidance from my advisor Professor Tim Holy. Work-ing with him exposed me to a wide range of science from olfaction, imagWork-ing to com-putation; without the time and effort he devoted into my study, I would not be able to complete this dissertation. I am especially grateful for his enthusiasm to share and carefully explain good things in work and even in life to us. I not only benefit from the specific skills and ideas he shares, but also learn an altitude that will positively influence me in my life. Thanks to Tim for giving me support as well as considerable freedom in research, which helps me learn to be independent and also preserve the original curiosity and imagination in science.

I am grateful to my thesis committee. They always gave me constructive sugges-tions in my update meetings, which helped consolidate my research data along the way towards publication. In the meantime their critiques helped me grow more mature in

(11)

An unexpected encounter with Professor O’Malley outside campus brought me plenty of extra insightful words. My lovely professors not only teach me in sciences, but also help be me become confident and independent in life.

My PhD study never got boring because of the accompany of all the members of Holy lab. Thanks to this large family. Thanks to everyone for giving me valuable suggestions. I also want to specially thank Diwakar Turaga, Ningdong Kang, Xiaoyan Fu for training me in special techniques, Kara Janiszak for taking care of my mice, and also Donghoon Lee who helped me treat mice when I was out of town. Particularly, two undergraduate students Mark Xu and Jack Chong helped me collect a portion of the behavioral data. Without their hard work on the nights in SIRF-west, I would not have accomplished everything presented in this dissertation. I also remember those days and weekends when I worked together with Zhongsheng Guo and Jian Wang to make the computer software control the newly developed microscope. On the new microscope, I am also collaborating with Maiko Kume in the mouse brain imaging. Thanks for their collaboration.

My acknowledgments also go to people of the WashU community. Yue Yang from Bonni lab and the members of the Kerschensteiner lab gave me valuable feedback when I presented my work in joint lab meetings. Tracy and Water at East McDonnell, and Eric and Wendy in SIRF-west provided superior technical support for my mice housed in the unconventional stacking-cages. John, Charlie and David in the machine shop made the custom parts for optics and mouse cages for my research. Dr. Kelly Monk gave me the HuC:GFP zebrafish for imaging. The Cavalli lab provided the space and technical help when I worked on ESP-1 purification. Dennis Oakley in the Bakewell neuroimaging

(12)

center was very helpful when I conducted confocal imaging during my rotation. I also want to thank my two rotation labs—DiAntonio lab and Milbrandt lab. Special thanks to Brad Miller for teaching me in my rotation study. Thanks to all these people who helped me. Without them, I will not have made it to today.

Finally, I would like to thank all my dear friends in St. Louis. There are so many names that I cannot list all of you, but I want you know that without your company, I would miss many beautiful moments in life. In the past years, several of them stayed so close to me throughout all my ups and downs. Thank Jie Zhang, Jiajin Xu, Haowei Yuan, Wan Shi for always being there and cheering me up everyday. Shuang Chen, Xiaoyan Fu and Xingxing Wang are the sweetest friends back me up for all situations. My strength and confidence also come from my two old friends Yuan Yuan in Texas and Wen Jin in California. Their care always arrives in time despite the distance. So does the love from my dear parents. Their unconditional love and trust for me never ends. If it is not their rearing and everlasting support, I will not grow up so happily, optimistic, and persistent. I love you all, my friends and family; thank you all!

(13)

ABSTRACT OF THE DISSERTATION

Large-scale Cellular Imaging of Neuronal Activity: a Study of Neural Individuality and

a Method for Imaging Mouse Cortex by

Pei Sabrina Xu

Doctor of Philosophy in Biology and Biomedical Sciences in Neurosciences

Washington University in St. Louis, 2014. Associate Professor Timothy E. Holy, Chair

The brain contains an enormous number of neurons with diverse gene expression, morphology, and connectivity. These neurons exhibit distinct activity in the course of behaviors. The study of neural coding of a specific behavior necessitates recording activ-ity from multifarious neurons in the circuit.One appealing approach is to simultaneously image the activity of a very large neuronal population at cellular resolution. However, recording calcium signals from tens of thousands of neurons at one time is not trivial. The gold standard technique, two-photon laser-scanning microscopy, typically permits recording from hundreds of neurons. Recently, we developed objective-coupled planar illumination (OCPI) microscopy, which uses thin sheets of light to image whole volumes of ∼ 10,000 neurons within 2 seconds. My dissertation includes an application and a further methodological development of such a fast large-scale imaging technique:

(14)

1) Large-scale functional imaging reveals individuality, dimorphism, and plas-ticity of mouse pheromone-sensing neurons. Different individuals exhibit dis-tinctive behaviors, which is presumably attributed to the neuronal differences between brains. However, studying neural individuality, especially at the level of the function of single neurons, requires an effective approach to measure cellular activity of a diverse neuronal population in a circuit. Here using OCPI microscopy, I performed calcium imaging of pheromone-sensing neurons in the intact mouse vomeronasal organ. Ex-haustive recording enabled robust detection of 17 functionally-defined neuronal types in each animal. Inter-animal differences were much larger than expected from random sampling, and different cell types showed distinct degrees of variability. One prominent difference was a neuronal type present in males and virtually absent in females, and animals exhibited a corresponding dimorphism in investigatory behavior. Surprisingly, this dimorphism was not innate but generated by plasticity, as exposure to female scents led to both the elimination of this cell type and alterations in behavior. The finding that an all-or-none dimorphism in neuronal types is controlled by experience—even in a sensory system devoted to “innate” responses—highlights the extraordinary role of “nurture” in neural individuality.

2) A new generation of OCPI microscopy enables unprecedentedlarge-scalein vivo

imaging of mouse brain activity by light-sheet microscopy. I have built a new variant of OCPI microscope, horizontal scanning objective-coupled planar illumination (hsOCPI) microscope, with enhanced imaging volume and speed by∼15 fold compared to OCPI, thereby capable of recording ∼ 100,000 neurons simultaneously. Using this technique, I imaged the entire nervous system of the larval zebrafish (including the

(15)

spinal cord) and a square-millimeter patch of mouse cortex ex vivo. The miniaturized optics around the specimen allowedin vivo imaging through a cranial window of a head-fixed mouse. This technique is the first application of light-sheet microscopy in calcium imaging of mouse cortexin vivo. The exceptional large-scale sampling of cortical activity with cellular resolution should usher new insights into the functions of brain circuits.

(16)

(17)

Chapter 1

Introduction

1.1 The motivation to record large-scale neural activity with cellular reso-lution

We are fascinated by the brain because it is the physical basis of our mind. It pro-cesses information we receive to form perceptions, sends commands to generate move-ment, stores our memories, harbors intelligence and emotion, etc. In order to understand how these neural events are created and unfold, a straightforward approach is looking into the brain to see what is there (structure) and what happens (activity).

1.1.1 An ongoing but urgent mission: recording neuronal activity

Knowing the brain structures provides the prerequisite to investigate brain activity. From the legendary Cajal drawings of brain cells with delicate arborizations, to the modern synaptic connectome of C. elegans [1, 2] and more recently of mouse miniature circuits [3, 4], our comprehension of the static brain structure promises to be accom-plished in only a matter of time. On the other hand, for each neuron in the brain, it

(18)

is the precise pattern of electricity coursing through it at a given time that determines what it is actually doing. Since the birth of the Hodgkin and Huxley model [5], electro-physiologists1 _{have made marvelous findings regarding neuronal activity, complemented}

with the growing imaging-based techniques. However, a major challenge still exists for the study of dynamic brain activity. It has been highlighted as a new Presidential focus, namely the BRAIN (Brain Research through Advancing Innovative Neurotechnologies) Initiative by the Obama government in February 2013.

1.1.2 Why we need recordings from large numbers of cells

The brain contains an enormous number of neurons. In the human brain, that number is at the level of a hundred billion. To examine the activities of those neurons, one can look at one neuron at a time or a population of neurons simultaneously. Now we are driven towards the latter direction, for several plausible reasons as listed below.

• Neurons do not operate individually. They are wired together — sequentially or laterally connecting to each other to form circuitry. During the course of a neural event, from the input to the output of the circuitry, there is a sequential neuronal activation pattern. Ideally, if one knows the precise timing, duration and amplitude of each individual neuron’s responses throughout the event, one can delineate the circuit design, understand the computation, and thereby ultimately decipher the “biological intelligence.” It is noteworthy to mention that a pure map of the static structure of the circuit is not sufficient to predict how the signal is flowing among 1_{To be precise, electrophysiology predates the birth Hodgkin-Huxley (HH) model. Galvani discovered} the neural electrical activity in the 1790s [6]. Another typical example is that Hartline performed the extracellular recording of individual retinal ganglion cells in the 1930s, which is still 20 years earlier

(19)

the neurons. The structure of synaptic/electrical connection does provide insight about the signal transmission locally at the connection, but not necessarily over multiple relay stages. In fact, the nonlinearity of the neural circuit introduced by the inhibitory connections makes that task almost impossible. A typical example is that after the C. elegans connectomic structure became available [2], there was still a need for functional connectomic study to map the activity pattern in order to understand how those 302 neurons work together.

To obtain the sequential activation pattern, in theory, one can record individual neurons one by one by repeating the same event multiple times, and then pool all the responsive neurons together and recapitulate the information flow according to the distinctive time delayed. However, even if one dared to set off that time-consuming adventure, there is another immediate challenge — the activation delays are at the level of millisecond or even smaller, and the variation of timing system in each recording will inevitably introduce inconsistency for the pooled neurons. The advantage of recording simultaneously is that systematic timing deviation will be consistent for all the neurons, and one-time recording of a large number of neurons is a much more efficient approach.

• Empirical evidence suggests that neuronal activities are not that “reproducible”. Although the trial-to-trial variation has been treated as “noise” for decades, emerg-ing awareness of the existence of waves, synchrony, and oscillation in circuits has changed the view. On the one hand, one can not deduce these features by examin-ing the activity of individual neurons; on the other hand, one-by-one measurement of many individual neurons will not detect these holistic phenomena that only

(20)

happen at particular times. Therefore, one has to gain the dynamic activity pat-terns of a sufficient number of neurons simultaneously to evaluate the population features.

• Neurons in the circuit are heterogeneous. The “type” of a given neuron can be defined by the genetic identity (gene expression), wiring identity (connection), and functional identity (response). It turns out that, no matter how the type is defined, diverse types, easily more than one hundred, co-exist in a local circuit, normally spatially distributed without an explicable pattern. All theses neurons are typically co-activated in the circuit, requiring cellular resolution to extract activity patterns of individually distinctive neurons. For a particular neural type, the initial discovery of the type necessitates a sufficient number of those particular neurons, which must come from an exhaustive sampling.

1.1.3 Potential recording approaches

During the past half century, the most feasible approach is recording the activity from a single neuron at a time, like whole cell recording, or up to tens of neurons using multi-electrode single unit recording. The recent imaging-based approaches like calcium imaging allow investigators to sample hundreds to thousands of neurons simultaneously. Other approaches like fMRI, as well as optical imaging based on slow, intrinsic signals, such as haemodynamic or light-scattering signals [7], are able to record large-scale pop-ulation signals, but they lack cellular resolution. Nowadays, electrode recording and fluorescent imaging approaches represent the most practical methods for cellular record-ing of neural activity. Focusrecord-ing on the theme “how many cells they can record,” each

(21)

technique will be introduced, and the strengths and limits of each technique regarding temporal/spatial resolution and potential for improvement will be also covered.

Electrode-based methods

The nature of the neuronal activity is the change of membrane potential, which gen-erates electrical signals capable of transmitting among cells in a circuit. This electrical activity was first discovered by Galvani over 200 years ago [6]. Since then, the natural way to record activity of neurons is by instruments capable of measuring the electric changes. Since the electrophysiologists were able to obtain reliable signals from each sin-gle neuron, they have been working in different ways to increase the number of neurons in a recording. Here are the two most commonly used techniques:

Whole-cell recording Whole-cell patch-clamp recording is currently the most pre-cise away to query intracellular ion activity of a neuron. The unprecedented sensitivity and temporal resolution allow one to study the “smallest” changes associated with neu-ronal activity, like the single ion channel action, quantal events of synaptic transmission (“minis”), etc. In patch-clamp experiments, each glass electrode is controlled by an individual manipulator, so the space around the tissue limits the number of electrodes that can be placed there. It is said that a 12-electrode setup can be managed, and the reported maximum number of successful simultaneous whole-cell recordings is 7-8 neurons [8]. Although robotic patch-clamping is being developed, due to the spatial constrains, it will be hard to increase that record further.

(22)

Single unit recording Extracellular recording of neurons normally detects the “com-pound” electrical activity of multiple neurons around the electrode 2_{. However, the}

mixed signal can be “separated” to extract signals from individual neurons. Basically, the more electrode probes, the more cells one can detect. To increase the number of probes, a multi-electrode array that arranges up to 256 electrodes laterally can be used for flat neural tissues. For deep neurons that require insertion of the shank-shaped elec-trode, one can increase the number of probes arranged in a single shank or even increase the number of shanks in the assembly. The most recent silicon electrode assembly con-tains 512 probes, which covered ∼1000 neurons within a radius of 140µm area in the rat cortex, and successfully recorded 163 neurons simultaneously [9]. One can infer the rough spatial distribution of all these neurons according to the probes’ position in the recording. One of the major limits is that the signals from adjacent neurons are normally impossible to obtain in this recording, because they are most likely “mixed” together in the same probes. On the other hand, the challenge to compact thousands of electrodes in a miniature device also restrains the further increase of the electrode number.

Temporal resolution of the electrode-based approaches is simply determined by the readout rate of electronics that can be easily achieved at 10KHz level. This sampling rate can sufficiently capture the dynamics of action potential that normally lasts for milliseconds. The merit of high temporal resolution makes electrode-based techniques the most favorable tool when timing information is of most concern, such as dynamic rhythm or noise analyses. However, the intrinsic limits on the cell number and lack of spatial information limit its potential in large-scale recording.

(23)

Fluorescence imaging-based methods

The nature of electrical activity has driven generations of neuroscientists to invest immense effort to build electric instruments to record changes in membrane potential. In the meantime, a minority of investigators also took advantage of voltage-sensitive dyes (optical membrane-voltage probes) to detect membrane potential changes by means of imaging. Similarly, the discovery of calcium fluctuation during the action potential and the availability of calcium fluorescent indicators, drove the rapid advancement of imaging approaches for recording neuronal activity. Nowadays many laboratories use fluorescence imaging-based methods to routinely measure the activity from tens and hundreds of neurons [10]. Combining the most state-of-the-art fluorescent probes with optical instruments with high speed and large field-of-view, a recent study achieved whole-brain imaging of neural activity of an organism [11–13].

Besides the remarkable advantages regarding the number of neurons, compared with electrode-based approaches, optical recording is minimally invasive. It avoids physical contact to neurons, though the potential for interference with cellular function due to calcium buffering. It is noteworthy to mention that the superior temporal resolution of electrode-based techniques is generally hard to beat. Although current functional indi-cators are fast and sufficiently sensitive to single action potentials, and fast optics allow sampling at millisecond resolution, yet that is attained with considerably compromised spatial capacity. In a typical population calcium imaging experiment, tens to hundreds of neurons are sampled every 2-3 seconds to follow the calcium fluctuation.

The detail of using fluorescent imaging approaches to measure neural activity will be introduced in the next section.

(24)

1.2 Large-scale fast fluorescent imaging

The implementation of fluorescent imaging of neuronal activity necessitates two ele-ments: a fluorescent indicator that reports activity of the cells, and a microscope that detects the fluorescence change. With the aim to record many neurons in a circuit at high speed over periods of minutes to hours, I will focus on the most advanced indicators and imaging technologies that meet, or have promise to meet the goal. Additionally, as the enormous growth of imaging volume (cell number) as well as recording time, the imaging data size has reached a level that greatly challenges our capability to handle big data, as briefly touched on at the end of this section.

1.2.1 Fluorescent indicators of neuronal activity

Back in the 1970s, two categories of chemical dyes were introduced to neuroscientists for monitoring neuronal activity. One is merocyanine dye that senses voltage changes [14]; the other one is a synthetic compound based on the well-known chelator EGTA, capable of detecting free calcium, namely BAPTA (1,2-bis(o-aminophenoxy)ethane-N,N,N’,N’-tetraacetic acid) [15]. Upon voltage change or calcium bindings, both dyes increase their quantum yield of fluorescence, thereby allowing quantitative3 imaging of neuronal activity.

Inspired by the chemical-based indicators, with the idea of fusing a voltage/calcium sensor with a fluorescence reporter, a series of protein-based indicators have been de-veloped, namely genetically encoded voltage indicator (GEVI) and genetically encoded 3_{For quantitative measurement, one would prefer sensors with a linear relationship between fluorescence} response and firing activity [16]. Nonlinearity in sensor can prevent precise spike quantification during action potential bursts [17].

(25)

calcium indicator (GECI). Many different calcium or voltage sensors have been dis-covered and used in designing new probes. As for the reporter, most of the GEVIs and GECIs adopt on of two strategies — either based on the single GFP’s fluorescence change, or through fluorescence resonance energy transfer (FRET) between two fluo-rescent molecules. One exception in the recently developed voltage indicator Arch, a microbial rhodopsine protein derivative, that itself is responsible for both voltage sensing and fluorescent emission.

For population cellular recording, currently the most commonly used probes are the calcium indicator GCaMP series. The GCaMP protein consists of a calcium-binding domain from calmodulin (CaM), a short linking region called M13, and a circularly per-muted enhanced GFP (cpEGFP). On calcium binding, the conformational changes in the calmodulin-M134 _{complex induce the fluorescence changes of GFP. Based on this same}

general structure, in the past decade, mutations at certain amino acids that changed the protein properties5_{, especially those that remarkably enhanced the extent of fluorescence}

change, produced different generations of GCaMPs. While larger fluorescence change normally means higher signal-to-noise ratio, a definite merit in physiological recording, along with that regard, there are multiple factors that need to be considered to choose the right indicators 6 _{for imaging neuronal activity at large-scale and cellular resolution.}

4_{M13 is the conformational actuator that is fused between CaM and GFP to enhance the conformational} change and induced fluorescent signal.

5_{Properties like calcium binding affinity and kinetics, the extent of binding induced protein} confor-mational change thereby fluorescence change, baseline GFP fluorescence level, dynamic range of the fluorescence etc..

6_{A detailed guide for GECI selection in different application has been reviewed by Hires et al. [18] and} Tian et al. [19].

(26)

Delivery A chemical-based indicator has to be experimentally delivered into the neu-rons right before the imaging, either by bulk loading (“multi-cell bolus loading”) [20], or facilitated by electroporation [21] or “gene-gun” [22]. It is technically challenging to get a large population of neurons effectively and relatively equally labeled in those procedures 7_{. On the contrary, a protein-based indicator can be introduced into the}

cells by transgenesis or viral constructs, so the labeling is much more efficient. Another advantage of genetic labeling is that it can target to specific neurons in the circuit that normally contain multifarious cells. At present, the Cre-dependent GCaMP transgenic mice and virus have been widely used in combination with existing Cre-lines [23], al-lowing specific recording from a variety of neuronal populations, and even within the specific subcellular compartments. It is worth noting that there are also studies that are beginning to look for strategy that allows facile targeting of dyes and drugs into specific cell classes [24, 25]. One of the ideas is screening fluorogenic ester resistant to endogeous esterases but hydrolysed by genetically controlled exogenous esterase [25].

Brightness In practice, in order to visualize responsive cells in the tissue, one needs a fluorescent probe that is sufficiently bright, which is related to the fluorescent quan-tum yield 8 _{of the indicator. The canonic fluorescent proteins GFP and its derivatives} 9 _{normally have high quantum yield} _∼₀_._{7, so those GECIs/GEVIs based on these}

flu-7_{Despite quite a few successes [20–22], more empirical data suggest that many cell types or tissues do} not take these dyes well, regarding penetration into the tissue as well as permeability across the cell membrane (even with the membrane permeable version).

8_{The definition of quantum yield is Φ =} #photons emitted

#photons absorbed. The actual brightness is determined by extinction coefficient×quantum yield.

9_{GFP derivatives are engineered GFP mutants that have different excitation/emission spectrums, such} as the yellow fluorescent protein (YFP) and cyan fluorescent protein (CFP) that are often employed for FRET-based GECI and GEVI. Red fluorescent proteins (RFP) like mCherry, is based on DsRed, a fruit fluorescent protein belonging to a different family to GFP. However, RFPs are erroneously referred to as GFP derivatives due to nomenclature convenience.

(27)

orescent proteins normally appear bright in the image. However, a recently developed GEVI, Arch [26], that is based on a microbial rhodopsin, has extremely low the fluores-cence quantum yield (∼10−3_{), therefore it is very challenging to detect its fluorescence}

signal.

Sensitivity In an imaging experiment, the neuronal activity is quantified as ∆F/F0,

where ∆F = Fobs −F0. Fobs is the probe fluorescence at a given time (normally peak

value during activation), and F0 is fluorescence at baseline. The signal-to-noise ratio

(SNR) is defined as the ratio of the fluorescence signal change 10_∆_F _{to the shot noise}

on the baseline fluorescence, F0N−1/2, where N is the number of photons detected [16].

Among the currently available indicators, calcium-dependent indicators generally have higher SNR than voltage-dependent indicators. Because of this, they are more commonly used in a variety of studies, especially in in vivo imaging in which animal movements introduce additional noises thereby decreasing the SNR.

Among GECIs, GCaMPs generally produce a larger fluorescence change than FRET-based indicators do, therefore they have higher SNR. GCaMP2 was developed in 2001, already able to robustly report spike trains. GCaMP3, developed in 2009, enhanced the fluorescence increase 4-6 fold [27], thereby further increased the SNR and the dynamic range. However, for a long time, GCaMPs could not beat the superior SNR of the chemical-based calcium indicators, such as OGB-1 AM, but that situation has been changed by the most recent GCaMP6. Nowadays, the fast version of GCaMP6 has been 10_{An equivalent calculation for FRET sensors takes the difference in donor-acceptor ratio divided by} the combined shot noise on the baseline fluorescence of both FPs.

(28)

used in in vivo recording to detect single action potentials in the neuronal somata, as well as synaptic calcium-transient in individual dendritic spines [28].

Photostability Continuously monitoring the activity requires an indicator that re-mains stable during repeated excitation. However, most indicators suffer from low photostability. Chemical-based voltage indicators are normally photobleached within seconds, while the recently developed GEVIs, including ArcLight [29, 30] and Arch [26], have improved lifetime. Compared to voltage indicators, calcium indicators generally exhibit much better photostability. Both the chemical version and GCaMPs (GCaMP2 and the higher versions) can be repeatedly imaged for hours without dramatic pho-tobleaching 11_{. Yet, GCaMPs are “chronically” expressed in the cells, which provides}

long-time window for imaging. One can repeatedly record the same cells over days and weeks to study the dynamic circuit during development or the process of learning.

Potential perturbation Both chemical- and protein-based indicators could poten-tially add calcium buffering or perturb the membrane electrical property, thereby affect-ing the functional integrity of the neurons [18, 31] and even the behavioral phenotypes [27]. At low probe concentration, it is difficult to visualize positive cells because of dim fluorescence; in contrast, highly-concentrated probes give increased baseline bright-ness and SNR, but can alter calcium homeostasis and compromise the signal. Thus, in both chemical loading and genetic expression, the abundance of the indicator needs to be controlled. Empirical evidence showed that genetically expressed indicator did 11_{It also depends on the imaging technique, for example confocal microscopy causes more serious} pho-tobleaching compared to two-photon or light sheet microscopy. See details in the next section.

(29)

not cause detectable abnormality [27, 32], which is probably because, in most cases, the genetically-expressed indicators present in cells with a relatively low concentration [33].

Voltage-sensitive dye is able to detect synaptic input and subthreshold activity, but its low signal-to-noise ratio and vulnerability to photobleaching make it less practical for repeated cellular recording. Most of the fluorescent protein-based voltage probes have been limited by small response magnitudes and slow kinetics. The rhodopsin-based new GEVI, Arch [26], is potentially attractive if its brightness can be increased. Currently, the start-of-the-art calcium indicators, specifically GCaMPs, are the preferred choices for large-scale cellular recording in the study of neural activity.

1.2.2 Fluorescence microscopy techniques for neural activity recording

To monitor the activity of a large population of neurons, the first two requirements on the optical instrument are: 1) large field of view to cover many neurons in the tissue, 2) high speed to follow fast biological processes such as Ca2+ _{or voltage dynamics. A}

wide-field fluorescence microscope may immediately comes to mind that directly takes full advantage of the advanced camera12with a large sensor and high speed. However, it only works well for imaging the surface of a tissue. When it focuses down to the interior, the signal normally deteriorates rapidly, mostly due to the emission light from out-of-focus region of the tissue. Therefore, the first key for volumetric imaging is to remove the out-of-focus light, and two general approaches have been used to do that. One 12_{It mainly includes two categories of cameras: charge-coupled device (CCD) and complementary} metal-oxide semi- conductor (CMOS) cameras.

(30)

is computationally subtracting the extraneous signals, and the other is instrumentally avoiding out-of-focus light by restricting the illumination, the so-called optical sectioning. The computational approach, known as deconvolution, is based on information about the process of image formation and recording to remove the out-of-focus light from each spot on the raw image. A different approach to drop the out-of-focus signal by com-putational procedure is structured illumination microscopy (SIM) [34, 35]. One of the properties of the out-of-focus signals is that they are insensitive to certain changes of illumination, for example, the three phase-shifted sinusoidal pattern [34]. After pair-wise subtraction of raw images obtained at different phases, common signals, mostly out-of-focus light, will be to some extent canceled out, but in-focus information that is modulated by the sinusoidal illumination pattern remains. This method achieves “op-tical sectioning” by taking multiple images with a conventional wide-field microscope, and has been successfully applied in three dimensions [35]. Alternatively, out-of-focus light can be assigned back to “non-focus” voxels in the three-dimensional space accord-ing to the angular information collected by additional microlenses [36]. Although these approaches can help remove/correct out-of-focus light computationally, they still leave behind the associated shot noise with wide-field illumination, as well as limited depth of penetration. For very thick samples with a large amount of out-of-focus light, they cannot produce images with sufficient contrast.

Classicoptical sectioning aims to deliver excitation light only to the focusing regions, that, in theory, can be a point, a line, a plane, or even an irregular pattern. Accord-ingly, a volumetric image in the three-dimensional space will be acquired one voxel, one line or one plane, at a time. In practice, the most popular optical sectioning technique

(31)

over past decades is point scanning microscopy, including confocal and two-photon mi-croscopy. However, the series point scanning is time-consuming, and many efforts have been made to improve their imaging speed. To date, for fast large-scale imaging, one may consider spinning-disk confocal microscopy and planar illumination-based so-called light sheet microscopy. Additionally, two-photon microscopy, as the most commonly used optical approach for recording neuronal activity, with exceptional advantage of beating scattering 13, and therefore is capable of penetrating deep into the brain, will also be introduced in this section.

Two-photon microscopy

Before explaining the imaging technique, I would first introduce the two-photon ef-fect. When using the standard green fluorescent fluorophore GFP for example, while one-photon effect describes the fluorophore molecule absorbing a single photon with wavelength of ∼ 488 nm and emit a photon of 515 nm, “two-photon excitation” is the quasi-simultaneous absorption of two near-infrared photons 14 _{and gives out a 515 nm}

photon. The probability for such a two-photon absorption event is low, and therefore requires high photon density, which is usually achieved using a high-energy pulsed laser and spatially focusing the laser beam through a high-numerical aperture (NA) micro-scope objective. Because that probability is in a quadratic relationship to the excitation 13_Scattering_{, is an essential optical property of most biological tissues, especially brain tissues [37]. Due} to the heterogeneity of refractive index of substances inside the tissue, light will not simply take the direct path but be diffracted, reflected and refracted by the tissue. This applies to both excitation and emission light in fluorescence imaging. Scattering causes the light redistribution to undesired regions, thereby resulting in out-of-focus illumination as well as lost of in-focus emission light (emission light from the in-focus regions appears to have been emitted from the last point of scattering but not from the actual location). This phenomenon gets worse as one images deeper into the tissue, substantially challenges the imaging technique.

(32)

intensity, successful excitation only takes place locally at the focal point, but not any non-focal regions. This is called “intrinsic optical sectioning”. Despite the light that is scattered inside the tissue, since all the emitted light is coming from the exact focus locus, one can collect as much as possible that is coming out of the tissue and assign all of them to the respective scanning point (origin). It doesn’t require a pinhole to eliminate the scattered light, and therefore has much higher photon budget compared with confocal microscopy. Along with the insusceptibility to scattering, longer excitation wavelength also allows better depth penetration. The attainable in vivo imaging depth has been reached to∼800−1000µm below mouse pia, the typical layer 5 neocortex [10, 38]. The insusceptibility to scattering also enhances the spatial resolution and signal contrast of the image. In addition, “intrinsic optical sectioning” confines effective illumination only to the vicinity of the focal plane, thereby largely reducing photobleaching/photoxicity of out-of-focus regions of the sample.

Due to the point laser-scanning nature of two-photon microscopy, collecting a vol-umetric image requires serially scanning the excitation point in the three-dimensional space, which is typically achieved by a pair of galvanometric mirrors for lateral scanning and a piezoelectric actuator for continuous axial movement 15 _{[40, 41]. While the speed}

of the scanning is somewhat limited by the attainable oscillation frequency of mechan-ics, the bottleneck of the imaging speed is the minimal dwell time required for each pixel. During the scanning, the laser has to dwell on each voxel for a period of at least 10−6_s 16 _{to collect sufficient photons thereby achieving reasonable signal-to-noise ratio}

15_{Novel approaches for fast z-scanning include remote mechanical z-scanning, temporal focusing, and} variable focus lenses, as reviewed by Lutcke et al. (2011) in “Two-photon imaging and analysis of neural network dynamics” [39].

16_{For detail mathematical reasoning, please refer to the supplementary information “Fundamental speed} limits of optical microscopy” in “Fast three-Dimensional fluorescence imaging of activity in neural

(33)

17_{. According to this minimal dwell time per pixel, and within a maximum of 2-3 s time}

window in calcium imaging, the conventional point scanning two-photon microscopy can scan 2−3×106 _{pixels. In a typical experiment, these pixels constitute a volume of}

256 pixels×256 pixels×40 frames, containing several hundred cells.

One way to overcome that speed/spatial limit is parallel illumination, where instead of focusing the laser into a single voxel, it splits the light into multiple “beamlets” (an array of beams) to illuminate multiple spots and image them with a camera at one time [44]. With a similar multi-focal strategy, combined with targeted path 18

approaches, several advanced two-photon microscopy techniques have been invented to increase imaging speed, in which a diffractive spatial light modulator (SLM) [45] or nonmechanical acousto-optic deflectors (AODs) [46] is used.

The multi-focal imaging strategy is to some extent going back to the “wide-field” fashion. The inevitable emission cross talk among multiple focal points would compro-mise the merit for overcoming scattering - the initial attraction of two-photon microscopy for deep brain imaging. However, it is always a balance among the three key criteria mentioned at the beginning of this section — imaging volume, speed and scattering. Dif-ferent from two-photon microscopy, spinning-disk confocal microscopy and the recently emerging light sheet technique balance the three criteria yet lean towards higher speed and larger volume.

populations by objective-coupled planar illumination microscopy” by Holekamp et al. (2008) [42]. Similar discussions can be found in chapter “Fluorophores for confocal microscopy” by Tsien et al. in “Handbook of biological confocal microscopy” by Pawley (2010) [43].

17_{See “signal-to-noise ratio” in page 12.}

18_{With pre-defined lines or discrete regions of interest, one can move the the laser focus only to the} selected pixels. Such targeted scanning avoids wasting time on background regions and maximizes signal integration time on the selected pixels (useful structures in images). With an acousto-optic deflector (AOD), the laser focus can be moved within microseconds between any two positions in a field of view.

(34)

Spinning-disk confocal microscopy

Similar to two-photon microscopy, conventional confocal microscopy is also a tech-nique based on point laser scanning. Except for the difference in wavelength, both deliver illumination light by the imaging objective that focuses the laser light into a point. However, without the two-photon effect, the excitation not only occurs at the desired in-focus point, but also extends to the non-focusing area that conforms to the point spread function of illumination light – a double cone excitation shape. In order to prevent out-of-focus emission light from non-focus planes, a pinhole is placed in the optically conjugated focal plane before the detector, and then the only signal collected is from the in-focus spot in each point scanning.

With the similar “multiplex” idea as mentioned for two-photon microscopy, in order to enhance scanning speed, different strategies have been used to illuminate multiple pixels simultaneously while collecting emission light from them at the same time. In practice, the most popular technique is spinning-disk confocal microscopy. It uses a disc with arrays of pinhole apertures that illuminate hundreds of spots simultaneously. During rapid rotation of the disc, the emitted fluorescence light from these spots is then focused onto either the same or different pinholes before forming images on a camera [47]. The arrangement and size of the pinholes on the disc allow scanning of the entire field of view during rotation. In addition, the same spots can even be illuminated several times within a frame time, thereby accumulating dwell time with reduced excitation energy. In fact, spinning-disk confocal normally uses an arc-discharge lamp for wide-field illumination, and the low excitation energy reduces photo-toxicity and photo-bleaching compared to the conventional confocal, making it the preferred system for live imaging.

(35)

However, the lower power illumination per pixel mentioned above on the other hand restricts the penetration depth. In addition, one major disadvantage of the spinning-disk confocal is its inefficient use of light. If pinholes occupy 1% of the disk area, 99% of the light is wasted. In order to increase the amount of light into each pinhole, a second disk containing an array of microlenses can be placed on top of the pinholes [48]. In such a design, a collimated laser beam is focused by those microlenses into the corresponding pinholes, which greatly increases the illumination efficiency.

In spinning-disk confocal, the pinhole aperture size, type and spacing affect the imag-ing speed; these parameters also determine the brightness of the out-of-focus excitation, thereby influencing the image quality. One would carefully choose these parameters to balance the image speed and quality according to specific needs. However, the configu-ration is normally fixed during manufacturing and cannot be adjusted by users.

Light sheet microscopy

In light sheet microscopy, all the pixels on the focal plane of the objective are illumi-nated and imaged at one time, so this technique possesses high speed and a large field of view, equivalent to that of wide-field microscopy. However, it limits the sheet illumi-nation to the vicinity of the focal plane, thereby minimizing out-of-focus light. The idea of planar illumination was first described by Siedentopf and Zsigmondy in 1903 [49], but it had not been used to image the interior of a biological tissue until ninety years later when it was implemented as a fluorescence light sheet microscopy [50].

(36)

How to create planar illumination? Planar illumination is not as widely under-stood as the single point illumination, which is normally produced by a regular objective lens in confocal or two-photon microscopy. In fact, in the latter techniques, it is the same objective lens that functions for both delivering point illumination as well as collecting emission light for imaging. In light sheet microscopy, a light sheet needs to be intro-duced from the side of the tissue to the focal plane of the imaging objective lens, so it requires a separate lens19. A light sheet can be generated by focusing the expanded laser beam in only one direction by a convex cylindrical lens or a uniaxial gradient-index lens [54, 55]. Alternatively, it can also be formed by a combination of a spherical lens and a concave cylindrical lens that spreads the light only in one direction. The latter combi-nation design allows the use of a conventional microscope objective lens as the spherical lens, therefore provides the flexibility to choose between different commercially avail-able objective lenses, and the convenience to change lenses through standard adapters. However, the low cost cylindrical/spherical lenses can be used to make miniaturized illumination assembly, the advantage of which will be discussed later.

Light sheet thickness Most light sheets are generated from Gaussian beams, the typical laser output. Because Gaussian beams undergo widening at increasing distances from the beam waist, the thickness of the sheet along the light propagation axis is uneven. The thinnest beam waist with thickness w0 is positioned at the center of the

field of view. The regions extended to the two sides of the beam waist where thickness 19_{Unconventional objectives incorporate the two parts in a single lens has been developed [51, 52]. The} designs make the illumination and even the detection on an oblique plane with a shallow angle to the focal plane of the objective lens, and therefore provide a very limited filed of view at the intersection of the two planes. Wolleschensky (2008) proposed a novel lens with an add-on cap to the existing microscope lens to achieve planar illumination [53], which has not been realized yet.

(37)

goes up to√2×w0 are defined as Rayleigh range, which is normally used as a reference

to determine the size of an acceptable field of view. The thickness of the sheet defines the axial resolution of a microscope, and the Rayleigh range affects the size of the field of view. Both the thickness w0 and the Rayleigh range are determined by the objective

numerical aperture (NA), and they counteract each other 20 _{Therefore, the larger the}

field of view, the worse (thicker) the axial resolution (light sheet thickness). In order to obtain a Rayleigh range of 100−500µm, a low NA∼0.1 lens is required, and the light sheet thickness will be approximately 4−10µm [42].

One way to generate light sheet with relatively even thickness is using a Bessel beam [56] instead of a Gaussian beam for illumination. This has been successfully implemented with a high NA lens to obtain a small-area ultra-thin sheet with 3D isotropic resolution down to∼0.3µm [57]. However, such a resolution improvement comes at a cost because the Bessel beams introduce “side lobes” of extraneous illumination outside the focal plane, thereby compromising the image quality and subjecting the sample to much more photobleaching and photodamage. Two practical approaches to relieve the consequences of the side lobes are structured illumination and two-photon excitation [57].

Microscope design and sample preparation As mentioned above, light sheet mi-croscopy uses two separate lenses, one for delivering excitation light sheet, and the other for collecting emission light. These two are orthogonally arranged in space. At the cross of the illumination and detection lens is the specimen. Sticking to this geometry 20_{For the planar gaussian beam, under the condition with an optimal NA for minimizing light sheet} thickness throughout the filed of view, there will bew{`}= 2λ`

πn, whereλis the light excitation

wave-length,nis the refractive index of the immersion medium. For details, please refer to the supplementary information “Upper limits on thickness of planar illumination” in “Fast Three-Dimensional Fluorescence Imaging of Activity in Neural Populations by Objective-Coupled Planar Illumination Microscopy” by Holekamp et al. (2007) [42].

(38)

arrangement, one can flexibly design a microscope with a preferred way to mount par-ticular samples. For example, light sheet can be delivered horizontally from the side of the sample, and the imaging objective is vertically mounted above the specimen. This example is a typical upright microscope, and many groups adopt it because of the conve-nience to build it in the context of a traditional wide-field epifluorescence microscope by adding extra components for light sheet illumination. There is another similar design, normally seen in a completely homemade version. In that design, the whole upright microscopy is laid down horizontally, and therefore everything can be mounted on the surface of an optical table. In both designs, a small specimen is normally placed in an imaging chamber with side “windows” allowing access of illumination/imaging lenses. However, for large samples like a head-fixed mouse with a cranial window on the top, neither design previously mentioned will work. In order to handle this most typical sam-ple preparation in the research of mammalian brains, Holy group invented a 45-degree vertically tilt design that allows both illumination and imaging from above [42]. The large tilt angle of light at the fluid/tissue interface normally introduces aberration 21

that to some extent can be alleviated by tuning the angle of the light sheet [54] as well as the adaptive imaging approach [60].

Scanning modality In high-speed neural imaging, 3D scanning is normally achieved by moving the rigid optics instead of the biological sample 22_{. During the scanning, the}

21_{It happens when the reflective indices of the immersion medium and sample are not precisely matched.} In a typical experiment with brain tissues, the tissue reflective index is normally 1.36−1.41 [58, 59], while the medium keeping the brain alive is normally aCSF with a refractive index of 1.34 [59]. 22_For_{in vivo} _{imaging, organism needs to be kept static to avoid undesired stimulation. Even for} _ex vivo preparation, high-frequency movement may induce tissue deformation due to inertial forth. In the study of developmental process in zebrafish or drosophila embryos, some investigators chose to move the specimen directly, probably because these specimens are not sensitive to movement.

(39)

light sheet and the imaging objective need to move in synchrony, which can be attained by synchronizing signals in the two separate actuators controlling the movement of the two parts, or physically coupling the two parts together [42] and moving them together with a single actuator. The latter design requires a special coupler to hold illumination optics together with the imaging objective, and this original invention is particularly named as objective-coupled planar illumination microscopy (OCPI).

Current constrains In light sheet microscopy, the large area parallel illumination allows one to take full advantage of the advanced high-speed megapixel camera (charge-coupled device (CCD) or complementary metal-oxide semi- conductor (CMOS) cameras) for fast volumetric imaging. Recently, the speed and image size has grown rapidly with the birth of high quantum efficiency, low-noise fast sCMOS camera. The state-of-the-art sCMOS with 2560×2160 pixels and 100Hz frame rate allows one gigapixel recording every 2 seconds, a calcium imaging time window. One recent application using such a fast camera was able to perform light sheet calcium imaging of 100,000 neurons of the entire larval zebrafish brain every 1.4 seconds [11]. With the anticipation of new generation higher speed cameras, it is crucial employ facile spatial scanning of a large area of the specimen. This is already starting to become a problem for large neural tissues like mouse brains, for which long travel of light delivery/collection optics is restricted due to the limited open space around the tissue. In addition, the travel speed and precision of the mechanics are also challenged in long-distance scanning. Recently the “scanless” electrically tunable lens has been adopted to achieve flexible volumetric imaging without moving the specimen or the objective, yet the light delivery optics still need to be moved during scanning. Another inherent limitation is the penetration depth. The two-photon

(40)

technique can improve the penetration, but the current light source cannot form a light sheet with high-density photons sufficient for two-photon excitation of the entire focal plane. While two-photon line scanning microscopy is achievable and yields images up to twice the depth of one-photon light sheet microscopy [61], it largely compromises the scanning speed. Either the microendoscopy approach that adds a microprism to relay the deep tissue imaging, or an in vivo tissue clearing technique that directly alleviates scattering could be the potential direction to improve the imaging penetration of light sheet microscopy.

1.2.3 Managing big data for large-scale imaging

Repetitive large-scale imaging generates large data sets. This is especially true for imaging awake behaving animals, in which wealth of stimuli can be presented to the animal while monitoring the activity of neurons in the brain. This type of experiment may last for hours, resulting in terabytes of data. Both online digital acquisition as well as offline processing and analysis of these data require considerable effort in the initial development of the platform. Without going deep into computer science and mathematic algorithms, here is a brief introduction of those basic aspects involved in practice.

Image acquisition software During a real time recording, big data poses challenges on continuous data transfer and storage. In order to attain high temporal resolution, high-performance imaging acquisition software must coordinate all the hardware and continuously write the data to hard drives with minimal overhead time. It is easier to employ the commercialized software coming along with the camera and incorporate it

(41)

into the framework, yet when one increases the imaging speed to the highest as claimed by the manufacture, poor performance my lead to “drop frames”. In this situation, customized optimization of the software pipeline may be required.

Image registration For hour-long recording, biological tissues may undergo morpho-logical changes like shifting, warping, swelling, etc.. In in vivo recording, there are motions caused by respiration or movements of restrained animals. In order to track the same neurons throughout the recording, imaging registration is needed to correct those mismatches of raw images. Algorithms that allow two-dimensional or three di-mensional, rigid or non-rigid corrections have been developed and utilized by different scientists. Which algorithms to choose depends on the nature of the data. Generally, a three-dimensional non-rigid correction is needed for warping tissues, such as ex vivo

preparation immersed in the medium. When the data set is so large that a single image volume reaches reaches one gigabyte, the optimization becomes a challenge for both the efficiency of the algorithm and the performance of the computer.

Image segmentation In order to extract the information of individual cells in the imaging volume, one needs to first determine which pixels are regions of interest (ROI), such as pixels representing cell bodies or neuronal processes. While manual image seg-mentation is possible for small size images, large images necessitate automatic algo-rithms for ROI recognition. There are two basic ideas to design the algorithm. The first one is based on anatomical criteria, which relies on the contrast of ROIs to the background. However, this approach becomes less practical as the new functional indi-cator like GCaMPs normally doesn’t provide sufficient contrast at baseline. Nowadays

(42)

the segmentation of ROI heavily relies on the functional criteria [62], as pixels that be-long to the same cells change intensity together, so temporal information will provide a useful basis for imaging segmentation. Different algorithms based on this idea have been developed and are still being improved upon [63]. It is worth pointing out that a good signal-to-noise ratio of individual neurons is required for robust ROI detection, as weakly responsive cells are normally difficult to recognize by algorithms. Addition-ally, for a heavily labeled, dense neuronal population, adjacent neurons that largely share similar response patterns normally require extra care to differentiate, for example by adding prior knowledge about the contour of ROIs, either by algorithm or manually. The latter strategy is called semi-automatic approach, which represents the main stream of the current status of the field.

1.3 Scope of the thesis

In the previous sections, I have reviewed approaches for recording activity from large number of neurons. Particularly, Light sheet fluorescence microscopy technique, with a relatively short history in neural imaging, is raising with potent momentum for fast large-scale neuronal recording. Focusing on this technique, specifically the objective-coupled planar illumination microscopy (OCPI) that was originally invented here by Holy group, my dissertation will further elaborate two directions of study.

A major body of this dissertation is an application of this technique in the research of neural individuality. This part of study was performed in the context of mouse pheromone-sensing neurons, which consist hundreds of distinct neuronal types. In or-der to investigate whether individual animals possess distinctive types of

(43)

pheromone-sensing neurons, I took advantage of OCPI imaging to exhaustively record activity from mice with different sexes and experiences to study potential sexual dimorphism and experience-dependent plasticity.

The second goal of this dissertation is to develop a new variant of OCPI microscopy to extend its application into in vivo imaging of mouse brain activity. The new technique, through a particularly designed horizontal scanning, combined with the advanced high-speed camera, also further enhances the imaging high-speed and scanning volume.

1.4 An application: neural individuality

What factors make individuals unique? Variability stems in part from genes and epi-genetics, and recent advances have permitted the study of such differences on a genome-wide scale [64]. However, much of what distinguishes individual persons or animals lies within the nervous system [65]. At a macroscopic level, inter-individual variability of the brain, for example in terms of cortical thickness/surface area [66] and grey/white matter structure [67], are correlated with performance in basic and higher cognitive functions [68, 69]. The emerging connectome [3] as well as single neuron innervation pattern anal-ysis [70] also suggest diverse neural wiring across individual organisms. Therefore, it is intriguing to study the individual differences in the nervous system as well as how the differences arise.

However, it is difficult to investigate individuality of neural structure and function. In a neuronal circuit, there are normally over 100,000 neurons of diverse types, intricate wirings, and synaptic/electrical connections. In order to make a comprehensive compar-ison between individuals, one has to sample a representative neuronal population from

(44)

each subject at the single-neuron level. As a result, there are extreme challenges on the resolution, scale as well as the efficiency of the research approaches. Large-scale cellu-lar imaging that simultaneously records activity from a very cellu-large neuronal population provides the opportunity to tackle this task. By exhaustive calcium imaging of a whole organism [11–13] or a local circuit [71, 72], one can examine individual differences in the functions of neurons.

1.4.1 Mouse pheromone-sensing neurons

We focused on the mouse vomeronasal epithelium, a component of the olfactory sys-tem devoted to social cues, often calledpheromones. Pheromones mediate a variety of in-nate social behaviors, including sexual behavior [73–82], male aggression [74–76, 83–86], interspecies defensive behaviors [87], maternal aggression [76–78, 80, 88] and lactating behavior [77]. These behaviors are specific to animals of particular sex, species, ages, experiences, etc.; it is very likely that neural circuitry for pheromone signaling con-tains information of animal identities. Pheromone-sensing neurons that directly detect pheromone chemicals and initiate the downstream signaling may be the fundamental source of individuality.

Mouse pheromone-sensing neurons, also called vomeronasal sensory neurons (VSNs), are packed in the vomeronasal epithelium(Fig. 1.1). Each sensory neuron sends an api-cal dendrite to the surface of the epithelium, where a receptor protein expressed at the dendritic tip can detect chemical cues from the nasal cavity. There are approximately 300 different vomeronsal receptor genes [91], and each sensory neuron mainly expresses a

(45)

Figure 1.1. Schematic of the structure of pheromone-sensing neurons. This represents a coronal section of the vomeronasal neuroepithelium. Each sensory neurons sends a dendrite to the surface of the epithelium to detect chemical cues. Apical neurons colored in magenta hues express receptor genes in the V1R family, while basal neurons in green hues express receptors in the V2R family. There is a small subset of basal neurons expressing formyl peptide receptor genes [89, 90] instead of V2R receptors. Courtesy of Timothy E. Holy.

(46)

single receptor type23. Due to the extensive receptor-type diversity and large number of neurons (over 100,000) in the vomeronasal epithelium, it is not trivial to compare these neurons between animals. Since the cloning of VSN receptors [89, 90, 93–96], molecu-lar approaches including in situ hybridization and microarray, as well as physiological recording with electrodes [97], have been used to study different types of VSNs. How-ever, these studies have been inconclusive [98], perhaps because the rarity of expression of each receptor gene (∼300 genes, each expressed by just 0.1–1% of the∼100,000 VSNs in each epithelium) [91] requires exhaustive sampling by methods sensitive to individual neurons. In this dissertation, I took advantage of the large-scale functional imaging approach to examine whether these VSN types vary by individual, by sex, and by expe-rience.

1.4.2 Sexual dimorphism

Neuronal individuality is generated by stochastic gene expression, biological factors (“nature”), and experience (“nurture”) [99, 100]. In wild-type mice, one prominent example of “nature”-induced individual variability is sexual dimorphism. Previous work has identified both sexually-dimorphic behaviors [92] and a small number of dimorphic nuclei or neuronal populations [101–109]. However, at present little is known about the physiological properties and circuit function of these neurons.

It is worth noting that in the mouse most dimorphic nuclei are downstream of mouse pheromone sensory neurons(Fig. 1.2) [92, 110, 111]. While the dimorphisms are well-established in “central” nuclei, it is unclear whether the dimorphism arises from the 23_{Exceptions exist in basal layer neurons expressing V2R receptor genes. These neurons express two} V2Rs, with a V2R2 subfamily gene as the most common co-receptor [92].

(47)

Figure 1.2. Schematic of the accessory olfactory system. Pheromone-sensing neurons situating in the vomeronasal organ (VNO) project to the accessory olfactory bulb (AOB), where the mitral cells send information to the deep brain nuclei (in purple), that are the previously identified sexually dimorphic brain regions: medial amygdala (A), ventromedial hypothalamus (Hyp), and bed nucleus stria terminalis (BNST). Redrawn and modified based on courtesy of Timothy E. Holy.

(48)

upstream sensory neurons. In several species, the male and female vomeronasal organs (VNOs), which house the pheromone-sensing neurons, are of different sizes [110,112,113]. However, none of the vomeronasal receptor genes have yet been found to be expressed differentially in male and female animals [94, 95, 114]. The closest known example is a difference, for cells expressing a particular V2R gene, of the mean soma depth within the VNO epithelium between the sexes [95]. However, because all VSNs receive sensory exposure through their dendritic knobs at the surface of the VNO, there is no known functional consequence of a shift in the location of the soma; moreover, this receptor gene is thought to be a pseudogene [98]. In addition, natural investigation of scents produces different amounts ofc-fos activation in VSNs of males and females [81,115,116]; however, because the amount of voluntary investigation time also differs by sex [117], it is unclear whether differences in c-fos activation reflect an underlying sex difference at the level of sensory neuron types and function.

1.4.3 Experience-dependent plasticity

Individual differences may result from differential experiences. It’s difficult to study the the contribution of experience in determining individual neural differences in humans due to the limited experimental approaches. Mice, which also display numerous plasticity in behaviors, offer the opportunity to pursue such questions at the level of neuronal circuitry. An extensive literature on cellular plasticity [118–123] provides a backdrop for individuality based on “nurture,” but how plasticity is organized on the scale of cell types, circuits, and systems remains largely mysterious.

(49)

In the case of mouse pheromone-sensing neurons, it has been shown that experience can change the VSN gene expression [114, 124]. These studie